Professional Documents
Culture Documents
A. A. Avilov: X Y G X
A. A. Avilov: X Y G X
A. A. Avilov: X Y G X
A. A. AVILOV
arXiv:2212.03653v1 [math.AG] 7 Dec 2022
Abstract. In this paper we classify nodal rational non-Q-factorial del Pezzo threefolds of
degree 2 which can be G-birationally rigid for some subgroup G ⊂ Aut(X).
1. Introduction
In this paper we work over an algebraically closed field k of characteristic 0. Recall that a
G-variety is a pair (X, ρ), where X is an algebraic variety and ρ : G → Aut(X) is an injective
homomorphism of groups. We say that G-variety X has GQ-factorial singularities if every
G-invariant Weil divisor of X is Q-Cartier.
Let X be a G-variety with at most GQ-factorial terminal singularities and π : X → Y be
a G-equivariant morphism. We call π a G-Mori fibration if π∗ OX = OY , dim X > dim Y , the
relative invariant Picard number ρG (X/Y ) is equal to 1 (in this case we say that G is minimal )
and the anticanonical class −KX is π-ample. If Y is a point then X is a GQ-Fano variety. If
in addition the anticanonical class is a Cartier divisor then X is a G-Fano variety.
Let X be arbitrary normal projective G-variety of dimension 3. Resolving the singularities
of X and running the G-equivariant minimal model program (see e.g. [31]) we reduce X either
to a G-variety with nef anticanonical class, or to a G-Mori fibration (see e.g. [23, §3]). So such
fibrations (and GQ-Fano varieties in particular) form a very important class in the birational
classification. In this paper we consider a certain class of G-Fano threefolds.
Definition 1.1. A projective n-dimensional variety X is a del Pezzo variety if it has at most
terminal Gorenstein singularities and the anticanonical class −KX is ample and divisible by
n − 1 in the Picard group Pic(X). If a G-Fano variety X is a del Pezzo variety, then we say
that X is a G-del Pezzo variety.
Del Pezzo varieties of arbitrary dimension were classified by T. Fujita ([18], [17], [19], see
also [22]). GQ-factorial G-minimal three-dimensional G-del Pezzo varieties were partially clas-
sified by Yu. Prokhorov in [25]. The main invariant of a del Pezzo threefold X is the degree
d = (− 12 KX )3 , it is an integer in the interval from 1 to 8. In this paper we consider the case
d = 2. In this case X is a double cover of P3 with ramification at a quartic surface. If d = 8
then X is a projective space. In this case equivariant birational geometry were studied by I.
Cheltsov and C. Shramov in the paper [13]. The cases d = 4 and d = 3 were considered in the
author’s previous works [2] and [3] (see also [6]). If d > 4 then X is smooth (cf. [25]) while
1
smooth del Pezzo threefolds and their automorphism groups are known well. For other types
of G-Fano threefolds there are only some partial results: see for example [27], [30].
Classification of finite subgroups of the Cremona group Cr3 (k) is one of the motivations
of this paper. The Cremona group Crn (k) is the group of birational automorphisms of the
projective space Pnk . Finite subgroups of Cr2 (k) were completely classified by I. Dolgachev and
V. Iskovskikh in [14]. The core of their method is the following. Let G be a finite subgroup
of Cr2 (k). The action of G can be regularized in the following sence: there exists a smooth
projective G-variety Z and a birational morphism Z → P2 which commutes with the action of
G. Then we apply the equivariant minimal model program to Z and obtain a G-Mori fibration
which is either a G-conic bundle over P1 , or a G-minimal del Pezzo surface. Dolgachev and
Iskovskikh classified all minimal subgroups in automorphism groups of del Pezzo surfaces and
conic bundles and so they obtained a complete list of finite subgroups of Cr2 (k). But quite
often two subgroups from such list are conjugate in Cr2 (k), so it is natural to identify them.
One can see that G-varieties Z1 and Z2 give us conjugate subgroups if and only if there exists
a G-equivariant birational map Z1 99K Z2 . So we need to classify all rational G-Mori fibrations
and birational maps between them as well.
Following this program in the 3-dimensional case one can reduce the question of classifica-
tion of all finite subgroups in Cr3 (k) to the question of classification of all rational GQ-Mori
fibrations and birational equivariant maps between them. Such program was realized in some
particular cases: simple non-abelian groups which can be embedded into Cr3 (C) (see [29], see
also [10], [12], [11], [9]) and p-elementary subgroups of Cr3 (C) (see [28], [26]).
Definition 1.2. By a node we mean an isolated ordinary double singular point of a variety of
arbitrary dimension. A variety is called nodal if its only singularities are nodes.
For applications to Cremona groups we are mostly interested in classification of rational del
Pezzo varieties. There is the result of I. Cheltsov, V. Przyjalkowski and C. Shramov which
states that almost all nodal non-Q-factorial quartic double solids are rational with one family
of exceptions (see [8, Theorem 1.5]). Also they conjectured that nodal Q-factorial quartic
double solids are not rational (see [8, Conjecture 1.9]). This is the reason why we consider only
non-Q-factorial quartic double solids here.
In this paper we are interested in the following problem: classify nodal non-Q-factorial
rational G-del Pezzo threefolds of degree 2 that have no G-equivariant birational map to a
“more simple” G-Fano threefold (for example P3 or a quadric in P4 ) or to a G-Mori fibration
with the base of positive dimension. In this paper we give a partial answer to this question.
The main result of this paper is the following theorem:
Theorem 1.3. Let X be a nodal non-Q-factorial del Pezzo threefold of degree 2 and G be a finite
subgroup of Aut(X) such that X is a rational G-Fano variety. Suppose that X is G-birationally
rigid. Then we have the following possibilities:
(i) X has 16 nodes. There is a 3-dimensional family of such varieties. All of them are
Aut(X)-birationally superrigid (see [5]).
(ii) X has 15 nodes. The variety of such type is unique and described in Proposition 4.1.
2
(iii) X has 12 nodes. There are 3 types and 6 families of such varieties (two 2-dimensional
and four 1-dimensional families). They are described in Propositions 7.8, 7.12, 8.8,
9.12, 9.14 and 9.15.
(iv) X has 10 nodes. There is a 1-dimensional family of such varieties. They are described
in Proposition 10.7.
(v) X has 8 nodes. There is a 2-dimensional family of varieties and one distinguished
variety. They are described in Propositions 12.7 and 12.8.
Remark 1.4. This theorem gives only necessary conditions for the variety X except for the
case of 15 nodes. We mostly don’t know if the variety X is G-birationally rigid in other cases.
Acknowledgements. The author thanks Yuri Prokhorov, Constantin Shramov and Andrey
Trepalin for useful discussions. This work is supported by Russian Science Foundation under
grant 18-11-00121.
G-del Pezzo threefolds were partially classified by Yu. Prokhorov in the paper [25] and the
next theorem is a part of its main result:
Theorem 2.1. Let X be a G-del Pezzo threefold of degree 2. There are the following possibili-
ties:
3
type r X Xb Z ∆0 ∆00 p s
15◦ 1 V2 − pt E7 − 0 10 − h
16◦ 2 (5.2.7) − P1 D6 A1 0 11 − h
17◦ 2 (5.2.2) − P2 A6 − 0 11
18◦ 2 (5.2.13) V3 pt E6 − 2 11 − h, h ≤ 5
19◦ 3 4.2.1 − (P1 )2 A5 A2 0 12
20◦ 3 (5.2.3) P2 A5 A1 2 12
21◦ 3 V4 pt D5 A1 4 12 − h, h ≤ 2
22◦ 4 (5.2.8) P1 D4 3A1 8 13 − h
23◦ 4 V5 pt A4 A2 6 13
24◦ 5 (5.2.5) P2 A3 A1 × A3 12 14
25◦ 6 8.1 V6 P2 A2 D5 20 15
26◦ 7 7.7 P3 pt A1 D6 32 16
Here r is the rank of the divisor class group, X/Z
b is a primitive birational model of the
0 00
variety X (see [25, Theorem 3.9]), ∆ and ∆ are root systems canonically associated with X,
p is the number of planes on X (see Definition 2.5 below), s is the number of singularities if
all of them are nodes and h := h1,2 (X),
b where X b is the standard resolution of X (for details
see [25, Introduction]).
We will consider all types of varieties in this table separately.
We are looking for G-birationally rigid varieties, and the following easy but very useful lemma
helps us to exclude many possibilities:
Lemma 2.2. Assume that the variety X is G-birationally rigid. Then the G-orbit of a singular
point cannot be of length 1, 2, 3 or 5. Also there are no G-invariant
e lines in P3 ⊃ Q.
Proof. Let p be a singular point of X. If p is a G-invariant singular point then the projection
from the image of this point in P3 gives us an equivariant link from X to a conic fibration.
If the G-orbit of p consists of two points, then the projection from the line passing through
their images gives us an equivariant link with a fibration by rational surfaces that can be
transformed into a Mori fibration with the base of positive dimension. The same is true if we
have a G-invariant
e line in P3 .
If the G-orbit of p consists of three points, then we have several cases. If their images in
3
P are collinear, then we again can apply the projection from the line passing through them.
If their images are not collinear, but the lines passing through pairs of this points lie on the
quartic Q, then the intersection of Q and the plane spanned by the orbit of p consists of four
lines, so we have a G-invariant
e line and again, thus, can apply the projection from it. If the lines
passing through pairs of this points don’t lie on the quartic Q, then X has a link to another
Mori fiber space by [4, Proposition 3.4].
If the G-orbit of p consists of five points, then these points are in general position. Indeed, if
they are lying on one plane, then the intersection of the quartic Q with this plane is a quartic
curve with 5 singular points of the same type. Then either this curve is a union of four lines
in general position, either it is a double quadric. In the first case the sixth singular point is a
4
G-invariant point, thus the group G cannot act transitively on five singular points (moreover,
one of them is G-invariant). In the second case there is a sixth G-invariant singular point of Q
lying on our double conic (see Proposition 2.6 below), so we can use the projection from this
point. So these points are in general position and we can consider the family of twisted cubics
passing through them. It is well-known that there is unique twisted cubic passing through six
points in general position. Since C · Q = 12, general twisted cubic intersects the quartic Q at
five nodes and two additional points. Thus the preimage of a general twisted cubic is a rational
curve and we have a structure of G-equivariant fibration by rational curves on X and it is not
G-birationally rigid.
Corollary 2.3. The group Aut(X)g cannot contain a cyclic subgroup of index 1 or 2. In
particular, it cannot be cyclic or dihedral.
Proof. If g is an element of finite order in PGL4 (k) then there is at least four g-fixed points in
P3 in general position. In any case we can find a couple of G-fixed
e points or a G-orbit
e of length
2, so we have a G-invariant
e line for any group G.
Remark 2.4. Assume that [p1 , p2 , p3 , p4 ] is a G-orbit of singular points in general position. We
may assume that their images in P3 have coordinates (1 : 0 : 0 : 0), (0 : 1 : 0 : 0), (0 : 0 : 1 : 0)
and (0 : 0 : 0 : 1). Then the equation of Q is of degree at most 2 in any variable. Thus we
have the birational map P(2, 1, 1, 1, 1) 99K P(2, 1, 1, 1, 1), that can be explicitly described by
the following formula:
(y : x0 : x1 : x2 : x3 ) −→ (yx0 x1 x2 x3 : x1 x2 x3 : x0 x2 x3 : x0 x1 x3 : x0 x1 x2 ).
One can easily check that in general such map transforms quartic double solid to another quartic
double solid. Thus, G-birational rigidity of X gives us a strong restrictions on coefficients of
the equation of Q. But if the equation of X is symmetric with respect to x0 , x1 , x2 , x3 then
automatically this transformation is a birational automorphism of X.
Definition 2.5. By a plane on a del Pezzo variety X we mean an irreducible surface Π ⊂ X
such that (− 21 KX )2 · Π = 1.
Proposition 2.6. Let H ⊂ P3 be a plane such that H ∩ Q is a double conic (we will call such
planes tropes). Then the preimage π −1 (H) is a couple of planes on X, and every plane on X
can be obtained in this way. Every plane on X contains exactly 6 nodes, and every couple of
planes contain exactly two common nodes.
Proof. The first statement is obvious. The second statement follows easily from a direct com-
putation in coordinates. Indeed, we may assume that H is given by the equation x0 = 0 and
Q has an equation of the form
F2 (x1 , x2 , x3 )2 + x0 G3 (x1 , x2 , x3 ) + x20 G2 (x1 , x2 , x3 ) + x30 G1 (x1 , x2 , x3 ) + ax40 = 0.
Singularities on H are given by x0 = F2 (x1 , x2 , x3 ) = G3 (x1 , x2 , x3 ) = 0 and there are exactly 6
of them (if the intersection of corresponding surfaces is not transversal at some point then this
point is more complicated singularity than node).
5
Corollary 2.7. The variety X of type 21◦ of Theorem 2.1 is never G-birationally rigid.
Proof. In this case we have exactly 2 tropes. The intersection of these tropes is an Aut(X)-
g
invariant line.
All other types we will study separately in the following sections.
3. Case 26◦
Varieties of this type was recently studied by I. Cheltsov. In the paper [5] he proved that all
such varieties are Aut(X)-birationally superrigid. The group Aut(X)
g is isomorphic to C42 o H
where H is a subgroup of S6 and PGL2 (k) which preserves some 6-tuple of points on P1 (see [5,
Lemma 10]). More precisely, he proved the following theorem:
Theorem 3.1. [5, Theorem 17] Let X be a quartic double solid with 16 nodes and G ⊂ Aut(X)
is a subgroup such that ClG (X) ' Z and C42 ⊂ G.
e Then X is G-birationally superrigid.
Remark 3.2. The condition C42 ⊂ Ge is not necessary, so the question of classification of all
subgroups G ⊂ Aut(X) such that the variety X is G-birationally rigid is still open.
4. Case 25◦
This case was completely studied in the previous author’s paper [4]. The main result of [4]
is the following
Proposition 4.1. The variety X is G-birationally rigid only in the following situation: it can
be given by the equation
4
X 4
X X4
y2 − 4 x4i + ( x2i )2 = xi = 0
i=0 i=0 i=0
in P(2, 1, 1, 1, 1, 1) and G is isomorphic to S5 × C2 , A5 × C2 or S5 (twisted subgroup in S5 × C2
which is not coincide with the first factor). Moreover, in this case X is G-birationally superrigid.
5. Case 24◦
In this case X has a small Q-factorialization X e which is the blow-up of a singular del Pezzo
threefold Y of degree 5 at three points p1 , p2 and p3 . The variety Y has divisor class group of
rank 2. Moreover, Y has exactly one node and fits into the following diagram
Yb / Yb +
f f+
{ ξ ~ ξ+ $
P2 = Z Y Z + = P1
where ξ, ξ + are small Q-factorializations, Yb 99K Yb + is a flop, f is a P1 -bundle and f + is a quadric
bundle (see [25, Theorem 5.2]; see also [20, Theorems 3.5 and 3.6] for detailed description of Yb
and Yb + ).
6
The variety X has 14 nodes and contains 12 planes, thus the quartic Q also has 14 nodes
and 6 tropes.
Lemma 5.1. The group Aut(X)
g acts on the configuration of singular points and tropes and
this action is faithful.
Proof. Obviously the group Aut(X)
g permutes singularities and tropes and preserves the relation
”a singularity lies on a trope”. If an element g ∈ Aut(X)
g preserves all singularities then there
3
is a plane in P containing 13 singular points or a line containing at least 7 singular points
(because the fixed locus of an element in PGL4 (k) is the projectivization of eigenspaces of a
corresponding element in GL4 (k)), which is impossible.
Let σ : Ye → Y be the blow-up of the unique singular point. Then Ye is a smooth Fano
threefold of Picard rank 3. Indeed, since −KYe = −σ ∗ (KY ) − E the anticanonical linear system
is the proper transform of the system of quadrics in P6 passing through the singular point
(here we consider the half-anticanonical embedding Y ⊂ P6 ). Obviously, such a linear system
separates points and tangent vectors at the singular point, so it is very ample.
One can easily calculate that (−KYe )3 = 38. According to [24] there are three families of
varieties with required properties: the blow-up of a smooth quartic at two points in general
position; the blow-up of a quartic at two skew lines; the blow-up of P1 × P2 at the smooth
curve C of bidegree (2, 1). In the first two cases there is no surface S ⊂ Ye that is isomorphic
to P1 × P1 and OS (S) ' O(−1, −1). In the last case S is a proper transform of the unique
divisor D of bidegree (0, 1) which contains C. Let E be the exceptional divisor of the blow-up
of C. Then pi ’s don’t lie on the proper transform of E, since otherwise the line on E passing
through pi has negative intersection with −KYe and this is impossible. So we have the following
sequence of morphisms
X ←− X e ←− Ye 0 −→ P1 × P2 ,
where the first morphism is the contraction of thirteen curves with zero intersection with
canonical class, the second morphism is the contraction of the proper transform of the divisor
D and the last morphism is the blow-up of the curve C and three points p0i .
Lemma 5.2. Singular points on X are images of proper transforms of the following subvarieties
in P1 × P2 :
• the divisor on P1 × P2 of bidegree (0, 1) that contains C;
• three curves of bidegree (1, 0) passing through p0i for some i;
• six curves of bidegree (0, 1) passing through p0i for some i that intersect C;
• three curves of bidegree (1, 1) passing through p0i and p0j for some different i and j that
intersect C;
• a curve of bidegree (1, 2) passing through p01 , p02 and p03 that intersect C.
Planes on X are proper transform of the following surfaces:
• three exceptional divisors of the blow-up of points p0i ;
• three divisors of bidegree (1, 0) passing through p0i for some i;
7
• six divisors of bidegree (0, 1) passing through p0i and p0j for some i and j which are
tangent to C.
Proof. One can easily check that subvarieties described above give us singular points and planes.
Since there are exactly 14 nodes and 12 planes, we have described all of them.
Corollary 5.3. Nodes of the quartic Q and tropes form a configuration that can be obtained
from the (154 , 106 )-configuration of nodes and tropes of a quartic with 15 nodes (see [4] for
detailes) by removing one node and four tropes that contain this node.
We denote such a configuration by (14, 66 )-configuration: every trope contains six nodes,
eight nodes lie on three tropes and remaining six nodes lie on two tropes.
Remark 5.4. One can visualize this configuration in the following way: tropes are faces of a
cube, nodes are vertices of a cube and centers of faces, every trope contains five nodes lying on
the corresponding face and the center of the opposite face.
Lemma 5.5. The group Aut(X)
g is a subgroup of S4 × C2 (the group of symmetries of the
cube).
Proof. From the previous remark we have the natural homomorphism from Aut(X) g to the
group of symmetries of a cube. Now we need to proof that this homomorphism is injec-
tive. Assume that the automorphism g ∈ Aut(X) g fixes four nodes corresponding to four
vertices of the cube lying on three intersecting edges. Without loss of generality we may
assume that these nodes have coordinates (1 : 0 : 0 : 0), (0 : 1 : 0 : 0), (0 : 0 : 1 : 0)
and (0 : 0 : 0 : 1). The action of the automorphism g is diagonal in this coordinates:
g(x0 : x1 : x2 : x3 ) = (α0 x0 : α1 x1 : α2 x2 : α3 x3 ). All tropes are g-invariant. Three of
them has an equation of the form xi = 0, i = 1, 2, 3, and they are g-invariant automatically.
Remaining three tropes has equations a10 x0 + a11 x1 + a12 x2 = 0, a20 x0 + a21 x1 + a23 x3 = 0
and a30 x0 + a32 x2 + a33 x3 = 0 and they are g-invariant if α0 = α1 = α2 , α0 = α1 = α3 and
α0 = α2 = α3 respectively. So all α’s are equal and g is trivial automorphism.
Lemma 5.6. The normal subgroup C32 ⊂ S4 × C2 acts on the set of tropes with three orbits of
length 2. A node lying on two tropes in one orbit doesn’t lie on a third trope. As a consequence,
if X is G-birationally rigid, then there is an Aut(X)-invariant
g triple of skew or coplanar lines
3
in P .
Proof. First two statements are easy observations. Thus, we have three C32 -invariant lines l1 , l2
and l3 (intersections of two tropes) that form an Aut(X)-invariant
g triple of lines. Since the
variety X is G-birationally rigid, we have no Aut(X)-invariant lines by Lemma 2.2, and there
g
are three possibilities: they are either skew, or coplanar, or intersect at one point.
If these lines intersect at one point then this point is common for all tropes and it is not
a node. We consider three tropes which has a common node. These tropes has two common
points and thus they have a common line and, consequently, have second common node which
is not our case (one can easily deduce this from the Corollary 5.2).
8
Lemma 5.7. If X is G-birationally rigid then lines lj cannot be skew.
where a 6= 0.
Proof. We know that the equation of the quartic Q restricted to P0 is a complete square and
points (1 : 0 : 0 : 0), (0 : 1 : 0 : 0), (0 : 0 : 1 : 0) and (0 : 0 : 0 : 1) don’t lie on Q. Without loss of
generality we may assume that coefficients of monomials x4i are equal to 1 (otherwise we can
apply a diagonal change of coordinates). Thus we may assume that
8. Case 20◦
In this case X has a small Q-factorialization X e which is the blow-up of a nodal cubic threefold
Y of Picard rank 2 in a general smooth point p. Such cubic has exactly 6 nodes and two
structures of P1 -bundle. By Lemma 2.2 the corresponding singular points of X form a single
G-orbit (let us denote them by p1 , p2 , p3 , p4 , p5 , p6 ). Remaining singular points of the variety
X come from six lines passing through the point p. Their images lie on the unique trope and
form a single G-orbit
e lying on a conic. Denote these trope and conic by P and C respectively.
14
Lemma 8.1. There is a G-invariant
e point q in P3 .
b ⊂ SL4 (k) be the preimage of the group G
Proof. Let G e ⊂ PGL4 (k). It is a finite group, and
the standard representation of G contains the 3-dimensional subrepresentation, corresponding
b
to the plane P . Thus it contains also 1-dimensional subrepresentation that corresponds to a
point q ∈
/ P.
There is a short exact sequence
0 −→ G0 −→ G
e −→ G00 −→ 0,
where the group G0 acts trivially on P and G00 acts faithfully on P .
Lemma 8.2. The group G0 is either trivial or cyclic of order 2.
Proof. If the group G0 is not trivial nor cyclic of order 2 then either all points pi lie on one line
passing through q (in this case G0 ' C6 ) or they lie on two lines passing through q (and in this
case G0 ' C3 ). In any case we have a G-invariant
e line: in the first case this is the line passing
through pi ’s, in the second case this is the intersection of the plane spanned by pi ’s and the
plane P .
Lemma 8.3. The group G00 is isomorphic to A4 or S4 .
Proof. The natural map G00 → PGL2 (k) is an embedding since every morphism acting trivially
on the curve C acts trivially on the plane P . Also we know that the group G00 has an orbit of
length 6 on the conic C. So G00 is isomorphic to C6 , S3 , D12 , A4 or S4 . But if G00 is isomorphic
to C6 , S3 or D12 then there is a G-invariant pair of points on C, so there is a G-invariant
e
line.
There are unique subgroups isomorphic to A4 or S4 in PGL3 (k) up to conjugation. They
can be constructed in the following way: if U is an irreducible 3-dimensional representation
of G00 ' A4 or S4 then P(U ) contains unique G00 -invariant conic C.
e The pair (P, C) with an
action of the group G00 is isomorphic to (P(U ), C).
e
Lemma 8.4. The projective space P3 with the action of the group G
e is the projectivization of
a 4-dimensional representation of the group G.
e
Proof. Let us construct such a representation. Let P3 = P(V ), P = P(U ) and q = P(W ),
where U, W ⊂ V are the corresponding vector subspaces of dimension 3 and 1 respectively.
As we noticed above we may consider U as the standard 3-dimensional representation of G00
and consequently of G.
e Then we can define in the unique way an action of G e on W and
consequently on V such that it agrees with an action on P3 . Obviously, this action is a linear
representation.
e is isomorphic to A4 , S4 , A4 × C2 , S4 × C2 or A4 o C4 .
Corollary 8.5. G
Proof. If G0 ' C2 then G
e is a central non-stem extension of G00 by C2 , since otherwise G0 acts
trivially on V .
15
Lemma 8.6. In some coordinates the quartic Q has a symmetric equation with respect to the
permutations of coordinates. As a consequence, Aut(X)
g ' S4 or S4 × C2 .
Proof. By Lemma 5.11 we may assume that A4 ⊂ G e acts on V by the permutations of coordi-
nates. If the equation of the quartic Q is not symmetric then it has the form
A(x30 x1 + x31 x0 + x32 x3 + x33 x2 + ξ(x30 x2 + x32 x0 + x31 x3 + x33 x1 ) + ξ 2 (x30 x3 + x33 x0 + x31 x2 +
+x32 x1 )) + B(x20 x1 x2 + x21 x0 x3 + x22 x0 x3 + x23 x1 x2 + ξ(x20 x2 x3 + x21 x2 x3 + x22 x0 x1 +
+x23 x0 x1 ) + ξ 2 (x20 x1 x3 + x21 x0 x2 + x22 x1 x3 + x23 x0 x2 )) + C(x20 x21 + x22 x23 + ξ(x20 x22 + x21 x23 )+
+ξ 2 (x20 x23 + x21 x22 )),
where ξ is a primitive root of the unity of degree 3. We also know that Q is singular at the
G-orbit of length 6, thus Q is singular at the point (a : a : b : b) for some a 6= ±b. Note that
we may apply the transformation
3 3 3 3
!
X X X X
(x0 : x1 : x2 : x3 ) 7−→ x0 + α xi : x1 + α xi : x2 + α xi : x3 + α xi ,
i=0 i=0 i=0 i=0
that corresponds to choosing another basic vector of the 1-dimensional subrepresentation W .
Thus we may assume that the quartic Q is singular at the point (1 : 1 : 0 : 0). Solving the
corresponding system of equations we see that C = −ξB, A = 2ξB. But the intersection of
such quartic with the plane P is not a double conic 2C.
Lemma 8.7. The variety X can be given by the equation
1
(2) s22 − s1 s3 + s21 s2 + A(8s1 s3 − 6s21 s2 + s41 ) = 0
4
3
xji . The group Aut(X) is isomorphic to S4 × C2 .
P
for some A 6= 1, where sj =
i=0
3
P
Proof. We know that the intersection of the variety X with the hyperplane xi = 0 is a double
i=0
conic. So the equation of the quartic Q has the form
(3) s22 + as1 s3 + bs21 s2 + cs41 = 0,
3
xji . As in the previous lemma we may assume that the quartic Q is singular at
P
where sj =
i=0
the point (1 : 1 : 0 : 0). Solving the corresponding system of equations we obtain the required
formula (2).
Assume that Aut(X)
g ' S4 × C2 . Since G0 acts nontrivially on W , C2 acts as − Id on W .
The generator of C2 acts as
3 3 3 3
!
1X 1X 1X 1X
(4) (x0 : x1 : x2 : x3 ) 7−→ x0 − xi : x1 − xi : x2 − xi : x3 − xi .
2 i=0 2 i=0 2 i=0 2 i=0
One can easily check that such a map doesn’t preserve the quartic Q.
16
From the previous lemmas we deduce the following proposition:
Proposition 8.8. The variety X can be given by the equation (2), the group Aut(X) ' S4 ×C2
and G ' S4 × C2 , S4 (which permutes two planes on X, otherwise X is not G-minimal) or
A4 × C2 .
9. Case 19◦
In this case X is a del Pezzo threefold with rk Cl(X) = 3 that contains no planes.
Theorem 9.1. (see [25, Theorem 4.2]) There are exactly six complete one-dimensional systems
of Weil divisors without fixed components on X. Each of them defines a structure of a quadric
fibration on some small Q-factorialization of X. They fits into the following graph
where two systems are connected by a line iff the corresponding quadric fibrations factors through
the same P1 -fibration.
Hence we have the following natural short exact sequence
0 −→ G0 −→ Aut(X) −→ G00 −→ 0
where G0 preserves all Fi ’s and G00 is a subgroup of D12 .
Remark 9.2. The Geiser involution acts on the hexagon as a central symmetry, because
F + τ (F) must be a multiple of − 21 KX .
Let F ∈ F1 be a general member and θ(F ) be the corresponding member of the linear system
−KX − F1 . Then F + θ(F ) is the preimage of a quadric under the double cover π. So systems
F1 and −KX −F1 give us a system of quadrics parametrized by a conic in the space of quadrics,
since a general point t ∈ P3 lies on two quadrics in our family (two points in the preimage π −1 (t)
lie in two different members of the linear system F1 ). We can explicitly write this system in
the form
(5) p2 Q1 + pqQ2 + q 2 Q3 = 0
where (p : q) parametrize a point in P1 and Qi = Qi (x0 : x1 : x2 : x3 ) are quadric forms. The
equation of the quartic Q in this case is simply the discriminant of (5)
Q22 − 4Q1 Q3 = 0.
17
Lemma 9.3. Three quadrics defined by Qi = 0 intersect transversally at 8 points (we will denote
quadrics by the same symbol as their equations). These points are nodes of the quartic Q.
Proof. Obviously, the common points of Qi ’s are singular points of Q, so there are finite number
of them. Let p be one of such points. If Qi is singular at the point p for some i then locally
the equation of Qi at this point contains no constant and linear terms, so the equation of the
quartic Q cannot contain the quadratic term of maximal rank. The same is true if tangent
planes to Qi ’s are not in general position.
Now we consider the equation (5) as a quadric form in variables xi whose coefficients are
quadric polynomials in variables p, q. Let M (p : q) be the corresponding symmetric matrix. Its
determinant is a polynomial of degree 8 (we denote this polynomial by D(p : q)), so in general
situation we have 8 singular members in the family. But in our case the situation is different.
Lemma 9.4. Let s be a node of the quartic Q which is not a common point of Qi ’s. Then this
point is the vertex of a singular member of the system (5) which corresponds to a root of the
polynomial D(p : q) of multiplicity at least 2. Moreover, if Q1 is this member then s ∈ Q2 .
Proof. Since s is a point of Q, it belongs to a single member of the system (5). We may
assume that it is the quadric Q1 , so (p : q) = (1 : 0). Since s ∈ Q and the equation of Q is
Q22 − 4Q1 Q3 = 0, we see, that s ∈ Q2 and s ∈ / Q3 . Thus if s is a smooth point of Q1 then it is a
smooth point of Q, because linear terms of Q and Q1 are proportional with non-zero coefficient
in a neighbourhood of s. Assume that Q1 is singular along a line. If Q1 is a double plane then
Q is singular along the whole curve Q1 ∩ Q2 . If Q1 is a pair of planes then the intersection of
one of these planes with Q is a double quadric, so the variety X contains planes and this is not
our case.
We may assume that s = (1 : 0 : 0 : 0). Since Q1 is singular at the point s, its equation
contains no terms x20 and x0 xi . Also we know that Q2 contains s, so its equation doesn’t
contain the term x20 . Hence the top left corner of the matrix M (p : q) is divisible by q 2 , while
all elements in the top row and left column are divisible by q. This entailed that D(p : q) is
divisible by q 2 , so (1 : 0) is a multiple root of this polynomial.
Corollary 9.5. D(p : q) has exactly four different roots of multiplicity 2, and vertices of the
corresponding quadrics are precisely 4 missing nodes of the quartic Q.
Lemma 9.6. The vertices of four singular quadrics are in general position.
Proof. Assume that three vertices are colinear. We may assume that quadrics Q1 , Q3 and
Q1 + Q2 + Q3 are singular at the points (1 : 0 : 0 : 0), (0 : 1 : 0 : 0) and (1 : 1 : 0 : 0) respectively.
One can solve the system of equations on coefficients which arise from this singularities and see
that all Qi ’s contain the line x0 = x1 = 0, so the quartic Q is singular along this line.
Assume that four vertices are coplanar. We may assume that quadrics Q1 , Q3 , Q1 + Q2 + Q3
and t2 Q1 + tQ2 + Q3 are singular at the points (1 : 0 : 0 : 0), (0 : 1 : 0 : 0), (0 : 0 : 1 : 0) and
(1 : 1 : 1 : 0) respectively. One can solve the system of equations on coefficients that arises
from this singularities and see that Q|x3 =0 is a double quadric, so the variety X contains planes
and this is not our case.
18
Assume that (1 : 0), (0 : 1), (1 : 1) and (t : 1) are roots of D(p : q) and (1 : 0 : 0 : 0),
(0 : 1 : 0 : 0), (0 : 0 : 1 : 0) and (0 : 0 : 0 : 1) are vertices of the corresponding quadrics. We will
call these points base nodes of the quartic Q. Using the fact that Q1 is singular at (1 : 0 : 0 : 0),
Q3 is singular at (0 : 1 : 0 : 0), Q2 contains (1 : 0 : 0 : 0) and (0 : 1 : 0 : 0), Q1 + Q2 + Q3
is singular at (0 : 0 : 1 : 0) and t2 Q1 + tQ2 + Q3 is singular at (0 : 0 : 0 : 1), we obtain the
following equations:
C32 o S4 (GAPId [192, 955]) for t = −1, 2 or 12 and C22 o S4 (GAPId [96, 227]) for
other values of t.
(v) The group Aut(X) is isomorphic to Aut(X)
g × C2 .
Proof. The first statement is obvious. The second statement follows from the previous lemmas.
One can calculate that the matrix
− β2 αβ
1 α
2 2 2
β
1
2α
1
2 2α
− β2
1 α
− 12 − α2
2β − 2β
1 1 1 1
2αβ
− 2β 2α 2
22
√ √
where α = −t and β = t − 1, preserves the equation of the quartic Q and obviously this
map does not belong to H, since H preserves the quadruple of base nodes. This proves the
third statement. The fourth statement can be easily proved with GAP using the action of the
group Aut(X)
g on the set of nodes.
For the proof of the fifth statement notice that matrices of the generators of the group Aut(X)
g
together with − Id generate the subgroup of GL4 (k) that is a double extension of the group
Aut(X).
g Thus we have an embedding of the group Aut(X)
g into an automorphism group of
P(2, 1, 1, 1, 1). Obviously the image of this embedding commutes with the Geiser involution, so
the group Aut(X) is the direct product of Aut(X)
g and C2 .
Proposition 9.13. Assume that H 0 ' C22 . Then we have the following:
(i) Exactly one of the numbers a12 , a13 , a23 , b02 , b03 , c01 is non-zero. We may assume that
a23 6= 0 without loss of generality.
(ii) The group H 00 is isomorphic to C22 (iff t = −1) or C2 (for other values of t).
(iii) The group H coincides with Aut(X).
g
(iv) We always have an Aut(X)-invariant pair of singular points so the variety X cannot
be G-birationally rigid.
Proof. The first statement is obvious. The second statement follows from the previous lemmas.
Assume that t = −1. We have Q2 = −2x22 + 2x23 = 2(x3 − x2 )(x3 + x2 ). Two planes
P1,2 = {x2 ± x3 = 0} contain six singular points of Q both (four common points of Qi ’s,
(1 : 0 : 0 : 0) and (0 : 1 : 0 : 0)). Quadruple of base nodes are distinguished from others
since they lie on even number of planes. If H is a subgroup of Aut(X)g of index 3 then we
have another such pair of planes, so there is a third plane P containing 6 singular points. This
plane contains at least 2 base nodes. Indeed, otherwise P and P1 or P2 contain at least three
common non-based nodes, but it is impossible. So the equation of P is of the form xi = 0 or
xi + αxj = 0. One can check that such a plane contains six nodes iff it is P1 or P2 . The case
t 6= −1 can be treated in the same way, but equations in this case looks more ugly.
Since H = G,
e (1 : 0 : 0 : 0) and (0 : 1 : 0 : 0) form an Aut(X)-invariant pair of points.
Proposition 9.14. Assume that H 0 ' C2 . Then we have the following:
(i) There are two cases :
(A) all coefficients in Sys(t; a12 , a13 , a23 , b02 , b03 , c01 ) with the same index are zero and
we may assume without loss of generality that b02 = b03 = c01 = 0;
(B) all coefficients are zero except two, such that the union of sets of their indices is
the set {0, 1, 2, 3} and we may assume without loss of generality that a23 and c01 6= 0.
(ii) In the case (A) the group√
H 00 can be isomorphic to C3 (we may assume without loss of
generality, that t = 1+2 3ı , a12 = a13 = a23 6= 0), C2 (we may assume without loss of
generality, that t = −1, a12 = a13 6= 0) or trivial.
(iii) In the case (B) the group H 00 is isomorphic to D8 iff t = −1 and c01 = ±a23 (we may
assume that c01 = a23 ), to C22 iff either t = −1 and c01 6= ±a23 or a23 = ±c01 with
arbitrary t 6= −1 (we may assume that a23 = c01 ) and to C2 in other cases.
23
(iv) In the case (A) we always have Aut(X)-invariant
g node or triple of nodes, thus X is
never G-birationally rigid.
(v) In the case (B) if c01 6= ±a23 or t 6= −1 then X is not G-birationally rigid.
2
√ (B) if c01 = a223 and t = −1 then Aut(X) is isomorphic to S4 × C2 if
(vi) In the case
c01 = ±2 3 and to D8 × C2 otherwise.
Proof. The first statement is obvious. The second and the third statement follows from previous
lemmas.
In the case (A) the point (1 : 0 : 0 : 0) is always H-invariant. If H ⊂ Aut(X)
g is a subgroup of
index 3 then we have either an Aut(X)-invariant
g singular point, or an Aut(X)-invariant
g triple
of points.
Assume that in the case (B) we have t = −1 and c01 6= ±a23 . In this case the group H is
generated by three maps
by direct computations that H acts on the set of singular points with three orbits of length
2, 2 and 8. In the first two cases H is a normal subgroup of Aut(X)
g of index 3, thus every
Aut(X)-orbit of singular points consists of one or three H-orbits of the same size. So we have
g
an Aut(X)-invariant
g pair of singular points. In the third case we can consider the normal
2
subgroup C2 ⊂ Aut(X). It acts with orbits of length either 1, 1, 2, 4, 4 or 2, 2, 4, 4. In both cases
g
the quadruple of base nodes is distinguished, thus H cannot be a subgroup of Aut(X) g of index
3. Thus we have that H = Aut(X) and there is a G-invariant pair of singular points.
g
Assume that in the case (B) we have t 6= −1 and c01 6= ±a23 . In this case the group H is
generated by two maps
and is isomorphic to C22 . One can easily check that τ preserves 4 singular points while σ
and τ σ don’t preserve any singular point. So if H ⊂ Aut(X)
g is a subgroup of degree 3 then
Aut(X) 6' A4 (since all elements of degree 2 in A4 are conjugate), so Aut(X)
g g is isomorphic to
D12 or C6 × C2 . In any case X is not G-birationally rigid.
24
Assume that in the case (B) we have a23 = c01 = c (we have no assumptions on t here). Then
we apply a change of coordinates given by the matrix
iβ αβ
2iδ α
αβγ
− 2α
− 2γ
− 2βδ
− iβ i −α −αβ
αγδ αβ βγδ 4δ
iβ i α αβ
− αγ αβδ βγδ 4δ 2
2i iβ αβ α
αβγδ
− 2αδ 2γδ 2βδ 2
√4
√ √ √ √
where α = tc2 − 4, β = −c + α2 , γ = c2 − 4, δ = 1 − t and = t. One can check
what we transform the equation of Q into the following:
4(tc2 + 4) 2 2 (c2 − 4)2 t2 2 2 (tc2 + 4)3
− 2 x0 x1 − 4x20 x22 − 4x20 x23 − 4x21 x22 − 2
x 1 x 3 + 2 2
x22 x23 +
c −4 4(t − 1) 4(c − 4)(t − 1)
√ √ √ √
8 1 − tc 2 2ct c2 − 4 2 8 1 − tc 2 2ct c2 − 4 2
√ x0 x2 x3 − √ x 1 x2 x3 − √ x2 x 0 x1 + √ x3 x0 x1 −
c2 − 4 1−t c2 − 4 1−t
2c2 (tc2 + 8t − 4)
− x0 x1 x2 x3 .
c2 − 4
2 −4)t
One can check that this equation corresponds to the quartic obtained from system Sys( (c4(1−t) ; 0,
q q
0, −2c c1−t2 −4 , 0, 0, −2c
1−t
c2 −4
). This system is isomorphic to the original system Sys(t; 0, 0, c, 0,
√
0, −c) iff c = ±2 2 − t. If t = −1 then the group H is a subgroup of Aut(X) g of index 3 and
is isomorphic to S4 × C2 , and similarly to Lemma 10.12 we have that Aut(X) ' S4 × C22 .
If t 6= −1 then the group H is a subgroup of Aut(X) g of index 2 which is not our case. We
also can apply the map (x0 : x1 : x2 : x3 ) 7→ (x1 : x0 : x2 : xt3 ) and obtain the system
q q
4(1−t) 1−t 1−t
Sys (c2 −4)t ; 0, 0, −2c c2 −4 , 0, 0, −2c c2 −4 , but this system is never isomorphic to the original
system. Obviously, all other permutations of coordinates lead either to one of these two systems
or to a system with a23 = 0 or c01 = 0, so in this case H is never a subgroup of Aut(X) g of
3
index 3. But if Aut(X) ' H ' C2 then there is an Aut(X)-invariant line.
g g
Proposition 9.15. Assume that H 0 is trivial. Then we have the following:
• H 00 can be trivial or isomorphic to C2 , C3 , C22 (two different cases), D8 or A4 .
• If H 00 is trivial or isomorphic to C2 , C3 , C22 or D8 then X is never G-birationally rigid.
√
• If H 00 is isomorphic to A4 (recall that in this case we may assume that t = 1+2 3i ,
√
a12 = a13 = a23 = −c01 = b02 = b03 t
) then Aut(X) ' A4 × C 6 if a 12 = 1 ± 3 and to
A4 × C2 otherwise.
Proof. If H 00 is trivial then the group Aut(X)
g is also trivial or isomorphic to C3 . In any case
we have an Aut(X)-invariant singular point or a triple of singular points. If H 00 ' C3 then
Aut(X)
g ' C3 , C9 of C23 . In any case we have an Aut(X)-invariant singular point or a triple
of singular points. If H 00 ' C2 then Aut(X)
g ' C2 , C6 or S3 . In any case X cannot be G-
birationally rigid.
25
Assume that H is isomorphic to C22 . If H = Aut(X) g then X is not G-birationally rigid, so
we need to check if H is a subgroup of Aut(X) of index 3. There are three groups of order 12
g
containing C22 : D12 , C6 × C2 and A4 . In the first two cases X cannot be G-birationally rigid,
so we need to check only the last case. If H acts on the set of base nodes with two orbits then
nontrivial elements of H cannot be conjugate (one can easily check that one of them acts on
the set of nodes as a product of 6 transpositions while two other elements act as a products of
5 transpositions), thus Aut(X)
g cannot be isomorphic to A4 . So we need to check only the case
2
of C2 acting on the set of base nodes transitively.
Let us consider the case then a12 = a 6= 0, b03 = −ta, b02 = a13 = b 6= 0, c01 = a23 = 0. In this
case the group H is generated by two maps
√ x0 √
x2
σ : (x0 : x1 : x2 : x3 ) 7−→ tx1 : − √ : tx3 : − √
t t
and √ √
√√
t − 1x2 tx1 x0
τ : (x0 : x1 : x2 : x3 ) 7−→ t t − 1x3 : − √ : −√ :√√ .
t t−1 t t−1
Maps σ, τ and στ act on P3 with two lines of fixed points which form a tetrahedron with vertices
√√ √ √ √√ √ √
p1 = ( t t − 1 : −i t − 1 : i t : 1), p2 = (− t t − 1 : i t − 1 : i t : 1),
√√ √ √ √√ √ √
p3 = ( t t − 1 : i t − 1 : −i t : 1), p4 = (− t t − 1 : −i t − 1 : −i t : 1).
Since Aut(X)
g ' A4 and C22 ⊂ A4 is a normal subgroup, the element of degree 3 should preserve
one point among pi ’s and permute cyclically three other points. Now we consider a map given
by the matrix √ √ √√ √√ √√
t√ t − 1 − √t t − 1 t√ t − 1 − √ t t−1
−i t − 1 i t√− 1 i t√ −1 −i t√− 1
√
i t i t −i t −i t
1 1 1 1
In new coordinates the equation of Q has the form
√ √ √ √ √ √
(ib − ta + 2 t − 1)2 x40 + (−ib + ta + 2 t − 1)2 x41 + (ib + ta + 2 t − 1)2 x42 +
√ √
+(−ib − ta + 2 t − 1)2 x43 + other terms.
√ √
Assume that ib − ta + 2 t − 1 = 0. Then one can check that an element of degree 3
which √acts like (x0 : x1 : x2 : x3 ) 7→ (x0 : αx2 : βx3 : γx1 ) preserve this equation only if
a = √2t−1t
, b = √−2i and the quartic Q is a double quadric, which is impossible. The same is
√ t−1 √ √ √ √ √
true if −ib + ta + 2 t − 1 = 0, ib + ta + 2 t − 1 = 0 or −ib − ta + 2 t − 1. If all of these
coefficients are nonzero we can apply a diagonal map and obtain an equation
x40 + x41 + x42 + x43 + 2x20 x21 + 2x22 x23 + other terms,
and the automorphism of degree three acts iff a2 + b2 = 4, but in this case Q again is a double
quadric. So this case is impossible.
26
Assume that b03 = ta12 , b02 = a13 , c01 = −a23 . In this case the group H is generated by two
maps
√ x0 √
x2
σ : (x0 : x1 : x2 : x3 ) 7−→ tx1 : √ : tx3 : √ and
t t
√ √
√√
t − 1x2 tx1 x0
τ : (x0 : x1 : x2 : x3 ) 7−→ t t − 1x3 : − √ : −√ :√√ .
t t−1 t t−1
Lines in P3 preserved by the group H form a family parametrized by the line P1 . The group
Aut(X)
g acts on this line with two fixed points, so there are two Aut(X)-invariant
g lines and X
is never G-birationally rigid.
Assume that H is isomorphic to D8 . If H = Aut(X) g then X is not G-birationally rigid, so we
need to check if the group H is a subgroup of Aut(X) of index 3. There are four groups of order
g
24 containing D8 : D24 , S4 , D8 × C3 and C3 o D8 . If Aut(X)g is isomorphic to D24 or D8 × C3
then X is not G-birationally rigid. So we need to check two other possibilities. We may assume
that t = −1 and we have two possibilities: either a12 = a13 = b02 = b03 = a 6= 0, a23 = c01 = 0
or a12 = a13 = b02 = −b03 = a 6= 0, a23 = −c01 = b.
In the first case we can explicitly calculate all singularities of X:
p1 = (1 : 0 : 0 : 0), p2 = (0 : 1 : 0 : 0), p3 = (0 : 0 : 1 : 0), p4 = (0 : 0 : 0 : 1),
p5 = (−a + α : −a + α : 1 : 1), p6 = (−a + α : −a − α : 1 : 1),
p7 = (−a − α : −a + α : 1 : 1), p8 = (−a − α : −a − α : 1 : 1),
√ √ √ √
p9 = ( 2 : − 2 : −1 : 1), p10 = (− 2 : 2 : −1 : 1),
p11 = (−a + α : −a + α : −1 − 2a(−a + α) : 1), p12 = (−a − α : −a − α : 2a(a + α) − 1 : 1),
√
where α = a2 − 2. Two generators of the group H act on the set of nodes as permutations
σ = (1, 4, 2, 3)(5, 8)(6, 12, 7, 11)(9, 10) and τ = (1, 2)(6, 7)(9, 10). If Aut(X)
g ' C3 o D8 then the
2
element σ generates its center, but it acts on the set of nodes as (1, 2)(3, 4)(6, 7)(11, 12), so we
have a distinguished quadruple of nodes which is impossible. If Aut(X) g ' S4 then where is a
2 2 2
normal subgroup σ ∈ C2 ⊂ D8 ⊂ S4 and all elements of order 2 in C2 are conjugate to each
other and thus have the same type of a cycle decomposition. But only σ 2 in D8 is a product of
4 transpositions, all other are products of 3 or 5 transpositions. So this case is also impossible.
In the second case we again have a 1-parametric family of H-invariant lines parametrized by
1
P , and even if H is a subgroup of Aut(X)g of index 3 then there is an Aut(X)-invariant
g line in
3
P. √
Assume that H is isomorphic to A4 , i.e. t = 1+i2 3 and a12 = a13 = a23 = b02 = b03 t
= −c01 = a.
Let us apply a map given by the matrix
β
α3 − α1 1
α tα
1 β
α α
− αi3 − α1
−α3 − 1 β 1
α α 2
t α
β
tα3 −α3 − α1 α
27
√
4
where α = t and p
a(t + 1) + a2 (t + 1)2 − 4t(a + 1)
β= .
2
One can check that this map transforms our quartic to a quartic with equation corresponding
to Sys(t; b, b, b, b, bt , −b), where
√
2a(8 − 8a + 4(3 − i 3)β
b= √ .
24a2 − 40a − 32 − 4(3 − i 3)aβ
This system is isomorphic
√ to the original one if a = b. One can check that this happens iff
a = −1 or 1 ± 3. But if a = −1 then the quadric Q1 is a union of two planes which is
impossible.
0 −→ G0 −→ Aut(X)
g −→ G00 −→ 0,
where the group G0 acts trivially on the P1 parametrizing the system of quadrics and G00 is
a subgroup of PGL2 (k).
Lemma 12.1. Either the group G0 is trivial or there are two skew lines such that both of them
contain four points vi .
Proof. Assume that the group G0 is nontrivial. Let g ∈ G0 be an arbitrary nontrivial element.
Then we have the following possibilities for the fixed locus Fix(g): (i) four points; (ii) two
points and a line; (iii) two skew lines; (iv) a point and a plane. The first case is impossible
since all points vi lie in Fix(g). The second case is impossible since in this case we have an
Aut(X)-invariant
g line.
In the fourth case the plane P contains seven or eight vi ’s. Consider the induced system of
quadrics p2 Q1 |P + pqQ2 |P + q 2 Q3 |P = 0. If its determinant is a nonzero polynomial then it
has degree 6, so this system contains at most six singular members, this is contradiction. If
the determinant is zero polynomial then the locus of degenerate conics in the linear system
αQ1 |P + βQ2 |P + γQ3 |P = 0 is a union of a conic and a line, so we have G00 -invariant conic or
pair of conics, thus G00 is a cyclic or dihedral group. Since G0 acts trivially on P (otherwise we
have too many vi ’s on one line and this line is G-invariant),
e we have G00 -invariant (and thus
Aut(X)-invariant)
g line on P .
In the third case both lines contain four vi ’s, otherwise they are Aut(X)-invariant.
g
Obviously, last two maps act on vi ’s as generators of the group D8 , so these three maps
generate Aut(X).
g One can check that they together with the Geiser involution generate the
group D8 × C22 .
Proposition 12.8. If the group G00 is isomorphic to S4 then there is unique variety with the
automorphism group S4 × C22 .
32
Proof. We may assume that (p1 : q1 ) = (1 : 0), (p2 : q2 ) = (α2 : 1), (p3 : q3 ) = (0 : 1),
1
(p4 : q4 ) = (α + 1 : 1), (p5 : q5 ) = (1 : 1), (p6 : q6 ) = ( 2−α : 1), (p7 : q7 ) = (α : 1),
(p8 : q8 ) = (β : 1), v1 = (1 : 0 : 0 : 0), v2 = (0 : 0 : 1 : 0), v3 = (0 : 1 : 0 : 0), v4 = (0 : 0 : 0 : 1),
v5 = (1√: 1 : 0 : 0), v6 = (0 : 0 : 1 : 1), v7 = (β : 1 : 0 : 0) and v8 = (0 : 0 : β : 1), where
α = 1±i2 3 , β = α1 . The group S4 generated by
σ : (p : q) 7−→ (α2 p + (1 − α2 )q : p − α2 q) and τ : (p : q) 7−→ ((α − 1)p + q : q)
acts transitively on the set of (pi : qi )’s. We have the induced action on vi ’s: σ transpose
v2i−1 with v2i and τ permutes them in the following way: v1 → v1 , v3 → v5 → v7 → v3 ,
v2 → v8 → v4 → v2 , v6 → v6 . One can solve the corresponding system of equations and obtain
the single solution up to multiplication by a scalar:
Q1 = x21 + (x3 − x2 )(x3 + βx2 ),
(11) Q2 = α2 x20 + x2 ((1 + α)x3 + αx2 ),
√
Q3 = −2 3ix20 − (1 + α)x21 + 2αx0 x1 + x22 − (1 + α)x23 + 2α2 x2 x3 .
One can easily see that the quartic Q is preserved by the map
(x0 : x1 : x2 : x3 ) 7→ (x0 : x1 : −x2 : −x3 )
which generates G0 . Also the equation of Q is preserved by the map
(x0 : x1 : x2 : x3 ) 7→ (x2 : x3 : x0 : x1 )
. More difficult but direct computation shows that it is preserved by the map given by the
matrix
1 −β 0 0
0 −β 0 0
0 0 β α
0 0 1 0
Obviously, last two maps act on vi ’s as generators τ, σ of the group S4 , so these three maps
generate Aut(X).
g One can check that they together with the Geiser involution generate the
2
group S4 × C2 .
References
[1] D. Abramovich, J. Wang. Equivariant resolution of singularities in characteristic 0, Math. Res. Letters 4
(1997), 427–433
[2] A. Avilov, Automorphisms of threefolds that can be represented as an intersection of two quadrics, Sb.
Math. 207:3 (2016), 315–330
[3] A. Avilov, Automorphisms of singular three-dimensional cubic hypersurfaces, Eur. J. Math. 4:3 (2018),
761–777
[4] A. Avilov, Biregular and birational geometry of quartic double solids with 15 nodes, Izv. Math., 83:3 (2019),
415–423
[5] I. Cheltsov, Kummer quartic double solids, ArXiv e-print 2202.11668
[6] I. Cheltsov, A. Dubouloz, T. Kishimoto. Toric G-solid Fano threefolds, ArXiv e-print 2007.14197
33
[7] I. Cheltsov, A. Kuznetsov, C. Shramov, Coble fourfold, S6 -invariant quartic threefolds, and Wiman-Edge
sextics, Alg. Num. Theory, 14:1 (2020), 213–274
[8] I. Cheltsov, V. Przyjalkowski, C. Shramov, Which quartic double solids are rational?, J. Alg. Geom., 28:2
(2019), 201–243
[9] I. Cheltsov, V. Przyjalkowski, C. Shramov, Quartic double solids with icosahedral symmetry, Eur. J. Math.
2:1 (2016), 96–119
[10] I. Cheltsov, C. Shramov. Five embeddings of one simple group, Trans. Amer. Math. Soc., 366:3 (2014),
1289–1331
[11] I. Cheltsov, C. Shramov, Cremona groups and the icosahedron, Monogr. Res. Notes Math., CRC Press,
Boca Raton, FL (2016)
[12] I. Cheltsov, C. Shramov. Three embeddings of the Klein simple group into the Cremona group of rank
three, Transform. Groups, 17:2 (2012), 303–350
[13] I. Cheltsov, C. Shramov. Finite collineation groups and birational rigidity. Selecta Math. (N.S.), 25:5, 71
(2019), 68 pp.
[14] I. Dolgachev, V. Iskovskikh, Finite subgroups of the plane Cremona group, in Algebra, arithmetic and
geometry: in honor of Yu. I.Manin. Vol. I, Progress in Math. 269, Birkhäuser Boston (2009), 443–548
[15] I. Dolgachev, Classical algebraic geometry. A modern view, Cambridge, Cambridge University Press (2012)
[16] I. Dolgachev, Abstract configurations in algebraic geometry, The Fano Conference, Univ. Torino, Turin
(2004), 423–462
[17] T. Fujita. On singular del Pezzo varieties. In Algebraic geometry (L’Aquila, 1988) Lecture Notes in Math.
1417 (1990), 117–128.
[18] T. Fujita. On the structure of polarized manifold with total deficiency one. I, II and III. J. Math. Soc.
Japan 32:4 (1980) 709–725, 33:3 (1981) 415–434, 36:1 (1984) 75–89.
[19] T. Fujita. Classification theories of polarized varieties. London Math. Soc. Lect. Notes Series 155 Cam.
Univ Press (1990), xiv+205 pp.
[20] P. Jahnke and Th. Peternell. Almost del Pezzo manifolds. Adv. Geom., 8(3):387–411, 2008
[21] Jong Hae Keum. Automorphisms of Jacobian Kummer surfaces, Compositio Math. 107 (1997), no. 3,
269–288
[22] A. Kuznetsov, Yu. Prokhorov. On higher-dimensional del Pezzo varieties, ArXiv e-print 2206.01549
[23] K. Matsuki. Introduction to the Mori program, Universitext, Springer-Verlag (2002), xxiv+478 pp.
[24] S. Mori, S. Mukai. Classification of Fano 3-folds with B2 ≥ 2. Manuscripta Math., 36(2):147–162, 1981/82
[25] Yu. Prokhorov, G-Fano threefolds, I, Adv. Geom. 13:3 (2013), 389–418
[26] Yu. Prokhorov. 2-elementary subgroups of the space Cremona group. In Automorphisms in Birational and
Affine Geometry, Springer Proc. Math. Stat., 79 (2014), 215–229
[27] Yu. Prokhorov. On G-Fano threefolds, Izv. Math., 79:4 (2015), 795–808
[28] Yu. Prokhorov. p-elementary subgroups of the Cremona group of rank 3, Classification of algebraic varieties,
EMS Ser. Congr. Rep., Eur. Math. Soc., Zurich (2011), 327–338
[29] Yu. Prokhorov. Simple finite subgroups of the Cremona group of rank 3, J. Algebraic Geom., 21:3 (2012),
563–600
[30] Yu. Prokhorov. Singular Fano manifolds of genus 12, Sb. Math. 207:7-8 (2016), 983–1009
[31] Yu. Prokhorov. Equivariant minimal model program. Russian Math. Surveys, 2021, 76:3, 461–542
[32] G. Sanna. Rational curves and instantons on the Fano threefold Y5 . PhD thesis, 2014
34